This paper investigates the cooperative tracking control problem for networked uncertain Lagrange systems with a leader-follower structure on digraphs. Since the leader’s information is only available to a portion of the followers, finite-time observers are designed to estimate the leader’s velocity. Based on the estimated velocity information and the universal approximation ability of fuzzy logic systems, a distributed adaptive fuzzy tracking control protocol is first proposed for the fault-free Lagrange systems. Then, the actuator faults are
considered and a distributed fault-tolerant controller is presented. Based on graph theory and Lyapunov theory, the convergence analyses for the proposed algorithms are provided. The
development in this paper is suitable for the general directed communication topology. Numerical simulation results are presented to show the closed-loop performance of the proposed control law
and illustrate its robustness to actuator faults and external disturbances.
1. Introduction
Recently, consensus problem for multiagent systems has attracted a great deal of attention in many fields such as biology, physics, robotics, and control engineering due to its broad applications in many areas such as mobile robots, unmanned air vehicles, autonomous underwater vehicles, satellites, aircrafts, automated highway systems, and air traffic control. In 1995, Vicsek et al. [1] proposed a simple but interesting discrete-time model of a system of several autonomous agents. It is shown that all agents eventually move in the same direction based on local information without any central controller or leaders. Jadbabaie et al. [2] provided a theoretical explanation for the consensus behavior of the Vicsek model by using the graph theory and the matrix theory. Chen et al. [3] investigated the finite-time distributed consensus problem for multiagent systems using a binary consensus protocol and the pinning control scheme. The readers are referred to [4–7] for more details.
Since a large class of networked mechanical systems can be modeled as a networked Euler-Lagrange system, it is fascinating to study the control problem of networked Euler-Lagrange system. The work in [8] studied the position synchronization problem for a group of Lagrange systems. However, all the robots required the desired common trajectory information. In [9], Ren presented some distributed consensus algorithms for the networked Lagrange systems, where only the undirected communication topologies were considered. The work in [10] proposed a consensus algorithm for networked leaderless Lagrange systems on undirected communication topology. The work in [11] investigated the finite-time cooperative tracking problem for a class of networked Euler-Lagrange systems with a leader-follower structure, where the communication topologies among the followers are undirected. The work in [12] studied the cooperative tracking control problem for a group of Lagrange vehicle systems with directed communication topology. The dynamics of the networked systems, as well as the target system, are all assumed unknown. A neural network (NN) is used at each node to approximate the distributed dynamics [12].
In reality, the actuators of the networked systems may undergo a partial loss of effectiveness or experience bias faults. Actuator faults often cause the control performance to deteriorate and even lead to catastrophic accidents. Thus, fault-tolerant control (FTC) schemes are proposed and used to guarantee system stability and acceptable performance. FTC design may be classified into two types: the passive approaches [13–15] and the active approaches [16–18]. The passive FTC approach designs a fixed controller that is able to tolerate only a limited range of predetermined faults, and, once implemented, it compensates for the anticipated faults without any online fault identification. However, the passive FTC has a very limited fault tolerance capability and is often designed to be conservative. The active FTC compensates for the effects of component faults by synthesizing a new controller online or by selecting a predesigned controller [19]. The active FTC requires a fault detection and diagnosis (FDD) mechanism to detect and identify the faults in real time, and then the controller is reconfigured based on the identified faults. Errors in fault detection may cause the controller to make wrong decisions. In this paper, we present a fault-tolerant control strategy which does not use any fault detection and isolation mechanism to detect, separate, and identify the actuator faults online. Although the fault-tolerant control methods have been used in a class of mechanical systems such as spacecraft [20] and near-space-vehicle [21], few works have focused on the fault-tolerant control of networked Euler-Lagrange systems.
Compared with the aforementioned works [2–12], this paper addresses the distributed fault-tolerant tracking problem for networked unknown Lagrange systems on digraphs. The distributed tracking control problem on digraph is more challenging than that on undirected or balanced graphs. We assume that the inertia matrix, the Coriolis/centrifugal matrix, the friction term, and the gravity term are all unknown for all Lagrange systems. A Lyapunov technique is utilized to design the distributed tracking controllers. By applying the universal approximation ability of fuzzy logic systems [22–24], an adaptive coordinated fuzzy controller is developed firstly when the actuators are fault-free. Then, an active fault-tolerant controller is developed, which can compensate for both the actuator bias faults and the loss of actuator effectiveness. The proposed control law does not require any FDD mechanism to detect the faults, which reduces the computation burden and decreases the response time of the controller. The analysis shows that if the designed nominal controller can ensure the stability of the fault-free distributed system, the proposed fault-tolerant controller guarantees the stability of the distributed system in the presence of faults. Numerical simulation results are given to show the effectiveness of the proposed method.
To our best knowledge, the fault-tolerant cooperative control of networked uncertain Lagrange systems on digraphs had not been fully investigated and it is still a challenging task. We present a solution to this problem in this paper. The main contributions of this research are described as follows.
A novel finite-time observer based cooperative control method is proposed for the networked nonlinear Lagrange systems. The dynamic leader and all the followers have unknown nonidentical dynamics. Compared with the results in [2–12], the proposed method is suitable for the general directed communication topology.
A robust fault-tolerant cooperative control scheme is presented to achieve distributed tracking control even if the actuator bias fault and the loss of actuator effectiveness coexist. In addition to the nominal controller, an auxiliary control input is designed to compensate for the actuator faults. It is proved that the proposed approach guarantees the convergence based on Lyapunov stability theory.
The remainder of this paper is organized as follows. Section 2 introduces the problem formulation and some preliminary results. Section 3 provides the proposed nominal controller and the robust fault-tolerant controller. The simulations are shown in Section 4. Section 5 concludes this paper.
2. Problem Formulation and Preliminaries2.1. Graph Theory
Let G=(V,ε,A) be a directed graph of order n, where V=[v1,…,vn] is the set of nodes, ε⊆V×V is the set of edges, and A=[aij] is called the adjacency matrix with weights aij>0 if (vj,vi)∈ε and aij=0 otherwise. We assume that the graph is simple; that is, (vi,vi)∉ε∀i, with no self-loops. Thus, aii=0. Define the in-degree of node vi as the ith row sum of A; that is, din(vi)=∑j=1naij. Define the diagonal in-degree matrix D=diag{din(vi)} and the graph Laplacian matrix L=D-A. The set of neighbors of a node vi is Ni={vj:(vj,vi)∈ε}, which is the set of nodes with edges incoming to vi. A directed path is a sequence of nodes v1,v2,…,vr such that (vi+1,vi)∈ε, i∈{1,2,…,r-1}. A directed tree is a directed graph, where every node, except the root, has exactly one parent. A spanning tree of a digraph is a directed tree that connects all the nodes of the graph.
2.2. Fuzzy Logic Systems
A fuzzy logic system (FLS) consists of four parts: the knowledge base, the fuzzifier, the fuzzy inference engine, and the defuzzifier. The knowledge base for FLS comprises a collection of fuzzy if-then rules of the following form:
(1)Rl:Ifx1isA1l,x2isA2l…,xnisAnl,ThenyisBl,l=1,2,…,M,
where x^=[x1,x2,…,xn]T∈U⊂Rn and y denote the FLS input and output, respectively. Ail, i=1,2,…,n and l=1,2,…,M, are fuzzy sets and Bl is the fuzzy singleton for the output in the lth rule. Fuzzy sets Ail and Bl are, respectively, associated with the membership functions μAil(xi)=exp(-((xi-ail)/bil)2) and μBl(yl)=1, where ail is the center of the receptive field and bil denotes the width of the Gaussian function. M is the rules number. By applying singleton function, center average defuzzification, and product inference, the FLS can be expressed as
(2)y(x)={∑l=1Myl(∏i=1nμAil(xi))}{∑l=1M(∏i=1nμAil(xi))},
where yl=maxy∈RμBl. Define the fuzzy basis functions as
(3)ξl(x)=[∏l=1nμAil(xi)][∑l=1M(∏i=1nμAil(xi))],
and let θT=[y1,y2,…,yl]=[θ1,θ2,…,θM] and ξ(x)=[ξ1(x),ξ2(x),…,ξM(x)]T. Then, the FLS (2) can be rewritten as
(4)y(x,θ)=θTξ(x).
Lemma 1 (see [23]).
Let f(x) be a continuous function defined on a compact set Ω. Then, for any constant ε>0, there exists an FLS (4) such that
(5)supx∈Ω|f(x)-θTξ(x)|≤ε.
By Lemma 1, FLSs are universal approximators; that is, they can approximate any smooth function on a compact space. Due to this approximation ability, we can assume that the nonlinear term f(x) is approximated as
(6)f(x,θ)=θTξ(x).
Define the optimal parameter vector θ*=argminθ∈Ωsupx∈U|f(x)-θTξ(x)|≤ε, where Ω and U are compact regions for θ and x, respectively. The FLS minimum approximation error is defined as
(7)ε=f(x)-θ*Tξ(x).
2.3. Problem Formulation
In this paper, the networked system consists of n+1 Lagrange systems, where the subsystem indexed by zero is assigned as the leader or target system and the other systems indexed by 1,2,…,n are referred to as the followers. The Euler-Lagrange equations of motion for the leader vehicle and the followers are described as
(8)M0(q0)q¨0+C0(q0,q˙0)q˙0+H0(q˙0)+G0(q0)=τ0,(9)Mi(qi)q¨i+Ci(qi,q˙i)q˙i+Hi(q˙i)+Gi(qi)=τi+δi,i=1,2,…,n,
respectively, where qi∈Rm(i=0,1,…,n) is the generalized configuration coordinate, Mi(qi)∈Rm×m is the inertia matrix, Ci(qi,q˙i)∈Rm×m is the Coriolis/centrifugal matrix, Hi(q˙i)∈Rm is the friction term, Gi(qi)∈Rm is the vector of gravitational torques, τi is the vector of control input torques, and δi represents the input disturbance. The inertia, Coriolis, friction, and gravity terms are all assumed unknown.
Note that the actuators of the Lagrange systems in (9) are assumed to be fault-free. They are called the nominal system. The case with actuator faults will be discussed in this paper. We will consider two types of actuator faults simultaneously, namely, the bias fault denoted by fi∈Rm and the loss of effectiveness of the actuators represented by a multiplicative matrix σi∈Rm×m. Hence, the dynamic model given by (9) can be rewritten as
(10)Mi(qi)q¨i+Ci(qi,q˙i)q˙i+Hi(q˙i)+Gi(qi)=σiτi+fi+δi.
Assumption 2.
For the bias fault, fi, there exists a positive continuous function ki(t) satisfying |fij|≤kij(t) for each j∈{1,…,m}. The actuator effectiveness matrix σi satisfies 0<σ-ij<σij≤1 (j∈{1,…,m}) for some constants σ-ij.
Assumption 3.
The input disturbance δi is bounded; that is, there exists a fixed bound δ0 such that ∥δi∥≤δ0.
Assumption 4.
The acceleration of the leader system is bounded by a positive scalar μ0 such that ∥q¨0(t)∥∞≤μ0<∞.
Property 1.
The inertia matrix Mi(qi) is symmetric and positive definite such that MmI≤Mi(qi)≤MMI for some constants Mm>0,MM>0.
Property 2.
The Coriolis/centrifugal matrix can always be selected so that the matrix (M˙i(qi)-2Ci(qi,q˙i)) is skew symmetric. Therefore, xT(M˙i(qi)-2Ci(qi,q˙i))x = 0 for all vectors x.
The distributed tracking control problem confronted here is as follows: we need to design the control protocols by using only local information such that the states of the follower systems synchronize to the states of the leader system; that is, one requires qi(t)→q0(t) and q˙i(t)→q˙0(t).
3. Main Results3.1. Finite-Time Observer Design
In this paper, we focus on the general case that the information of the leader system is only available to a portion of the followers. Observer based control methods are used in this paper; that is, finite-time observers are used to estimate the leader’s information. We first design a velocity observer for each follower as follows:
(11)z˙i=-γsgn{∑j∈Niaij(zi-zj)+bi(zi-q˙0)},i=1,2,…,n,
where γ>μ0>0 is a positive constant and zi denotes the estimate of q˙0 on the ith Lagrange system.
Let D+V(t,x) be the upper right-hand Dini derivative of V(t,x) with respect to t; that is,
(12)D+V(t,x)=limsuph→0+V(t+h,x(t+h))-V(t,x(t))h.
Lemma 5 (see [25]).
Let I0={1,2,…,n} and let, for each i∈I0, Vi(t,x):R×Rm→R (i=1,…,n) be c1. Define V(t,x)=maxi∈I0Vi(t,x). Let I(t)={i∈I0:Vi(t,x)=V(t,x)} be the set of indices where the maximum is reached at time t. Then D+V(t,x) satisfies D+V(t,x)=maxi∈I(t)V˙i(t,x).
Lemma 6.
Let the communication graph G contain a spanning tree with the root node being the leader system (8). For the observers (11), it holds that zi→q˙0(t) (i=1,…,n) in finite time. The finite setting time T0 satisfies
(13)T0≤maxk{maxj(zjk(0))-minj(zjk(0))}γ-μ0,
with k∈{1,…,m} and j∈{0,1,…,n}.
Proof.
Let z0=q˙0 and ai0=bi. Then, (11) can be rewritten as
(14)z˙ik=-γsign{∑j=0naij(zik-zjk)},i=1,…,n,k=1,…,m.
Define eik=zik-q˙0k. Thus,
(15)e˙ik=-γsign{∑j=0naij(eik-ejk)}-q¨0k.
Choose the Lyapunov function
(16)Vk=maxi{eik}-mini{eik}.
It is obvious that Vk is a positive definite, Lipschitz continuous, and regular functions. Let Nmax denote the set of indices i for which eik=max{eik:i∈{0,1,…,n}}. Let Nmin denote the set of indices i for which eik=min{eik:i∈{0,1,…,n}}. In light of Lemma 5, one has
(17)D+Vk=maxi∈Nmax{e˙ik}-mini∈Nmin{e˙ik}.
If D+Vk=0, then maxi∈Nmax{eik}=mini∈Nmin{eik}; that is, zik=q˙0k. If D+Vk>0, then Nmax≠ϕ and Nmin≠ϕ. Since the communication graph G contains a spanning tree, there must exist a node l∈{0,1,…,n}/Nmax (or l∈{0,1,…,n}/Nmin) such that a path from l to the node in Nmax (or the node in Nmin) exists; that is,
(18)maxi∈Nmax{e˙ik}=-γ-q¨0k,
or
(19)mini∈Nmin{e˙ik}=γ-q¨0k.
Thus,
(20)D+Vk≤-γ-q¨0k≤-γ+|q¨0k|≤-(γ-μ0).
Integrating both sides of the above equation between the time t=0 and t=T0 yields
(21)T0≤Vk(0)γ-μ0≤maxk{maxj(zjk(0))-minj(zjk(0))}γ-μ0.
Thus the claim follows.
3.2. Nominal Controller Design for the Fault-Free Lagrange SystemsDefinition 7.
The local neighborhood position errors for subsystem i are defined as
(22)si=∑j∈Niaij(qi-qj)+bi(qi-q0).
Definition 8 (see [12]).
The position and velocity tracking errors for each Lagrange system are said to be cooperatively uniformly ultimately bounded if there exist compact sets Φ1∈Rm and Φ2∈Rm such that, for all qi(t0)∈Φ1 and q˙i(t0)∈Φ2, there exist bounds B1 and B2 and a time T(B1,B2,qi(t0),q˙i(t0)) such that ∥qi(t)-q0(t)∥≤B1 and ∥q˙i(t)-q˙0(t)∥≤B2 for all t≥t0+T.
Define the sliding-mode error
(23)ei=q˙i-zi+si.
For the follower system i, the error dynamics are given by
(24)Mie˙i+Ciei=Miq¨i-Miz˙i+Mis˙i+Ciei.
In light of (9), one has
(25)Mie˙i+Ciei=(τi+δi)+Fi(xi)-Miz˙i,
where
(26)Fi(xi)=-Gi-Hi-Ciq˙i+Mis˙i+Ciei.
The function Fi(xi) is unknown. In light of the approximation property of fuzzy logic systems, there exist weights θi such that
(27)Fi(xi)=θiTξi(xi)+εi.
For further analysis, we need the following assumptions.
Assumption 9.
On a compact set Γ, the fuzzy logic system approximation error εi in (27) is bounded; that is, there exists a fixed constant bound εi0 such that ∥εi∥≤εi0.
Assumption 10.
There exists a constant θi0 such that the weight matrix θi is bounded by ∥θi∥F≤θ0, ∀i.
Now we get one of the main results.
Theorem 11.
Consider the networked Lagrange systems (8) and (9). Assume that the directed communication graph G has a spanning tree. Make Assumptions 3–10. Take the distributed adaptive controller as
(28)τi=-Kiei-θ^iTξi-3ei4μi,
with Ki=kiI,(ki>0) and μi>0, and the distributed adaptive tuning law as
(29)θ^˙i=βiξieiT-ηiθ^i
with βi>0 and ηi>0. The position and velocity tracking errors are cooperatively uniformly ultimately bounded (UUB). The ultimate bounds can be made as small as possible.
Proof.
Consider the candidate Lyapunov function
(30)V=12∑i=1neiTMiei+∑i=1n12βitr{θ~iTθ~i},
with θ~i=θi-θ^i. Taking the time derivative on both sides of (30) yields
(31)V˙=∑i=1n(eiTMi(q¨i-z˙i+s˙i)+12eiTM˙iei)-∑i=1n1βitr{θ~iTθ^˙i}.
By substituting (9) into (31), we have
(32)V˙=∑i=1n(+12eiTM˙ieieiTMi(Mi-1(τi+δi-Gi-Hi-Ciq˙i)-z˙i+s˙i)+12eiTM˙iei)-∑i=1n1βitr{θ~iTθ^˙i}=∑i=1n(+12eiTM˙ieieiT(τi+δi-Gi-Hi-Ciq˙i-Miz˙i+Mis˙i)+12eiTM˙iei)-∑i=1n1βitr{θ~iTθ^˙i}.
In order to apply Property 2, we first add the term eiTCiei and then subtract it. Thus
(33)V˙=∑i=1n(eiT(τi+δi-Gi-Hi-Ciq˙i-Miz˙i+Mis˙i)+eiTCiei)-∑i=1n1βitr{θ~iTθ^˙i}.
Define
(34)Fi(xi)=-Gi-Hi-Ciq˙i+Mis˙i+Ciei.
According to (27), we have
(35)V˙=∑i=1neiT(τi+θiTξi(xi)+εi+δi-Miz˙i)-∑i=1n1βitr{θ~iTθ^˙i}.
In light of Young’s inequality, one has
(36)eiTεi+eiTδi-eiTMiz˙i≤∥ei∥(∥εi∥+∥δi∥+∥Miz˙i∥)≤3∥ei∥24μi+μi∥δi∥2+μi∥εi∥2+μi∥Miz˙i∥2≤3∥ei∥24μi+μi(δi02+εi02+mγ2MM2).
Substituting (28) and (36) into (35) yields
(37)V˙≤∑i=1neiT(-Kiei+θ~iTξi(xi))-∑i=1n1βitr{θ~iTθ^˙i}+∑i=1nμi(δi02+εi02+mγ2MM2).
From (29), we have
(38)V˙≤-∑i=1neiTKiei+∑i=1nηiβitr{θ~iTθ^i}+∑i=1nμi(δi02+εi02+mγ2MM2).
Note that
(39)tr{θ~iTθ^i}≤∥θ~i∥F∥θi∥F-∥θ~i∥F2≤-12(∥θ~i∥F2-∥θi∥F2).
Thus,
(40)V˙≤-∑i=1n(kieiTei+ηi2βitr{θ~iTθ~i})+∑i=1n(μi(δi02+εi02+mγ2MM2)+ηi2βi∥θi∥F2).
Let d0=∑i=1n(μi(δi02+εi02+mγ2MM2)+(ηi/2βi)∥θi∥F2). Thus,
(41)V˙≤-∑i=1n(kiMmeiTMiei+ηi2βitr{θ~iTθ~i})+d0.
Moreover,
(42)V˙≤-c0V+d0
with c0=min{(2ki/Mm),ηi}.
Let e=(e1T,e2T,…,enT)T. From (42), one has
(43)∥e∥2≤2V(0)Mme-c0t+2d0Mmc0.
with Mm defined in Property 1.
Note that zi=q˙0 as t>T0. Thus, we have
(44)S˙=-((L+B)⊗Im)S+((L+B)⊗Im)e,t>T0,
with S=(s1T,s2T,…,snT)T. Since L+B is a positive stable matrix, there exists a symmetric positive matrix P0 such that Q=P0(L+B)+(L+B)TP0 is symmetric positive definite. Choose the Lyapunov function
(45)V1=ST(P0⊗Im)S.
The time derivative of V1 is
(46)V˙1=2ST(P0⊗Im)(-((L+B)⊗Im)S+((L+B)⊗Im)e)=ST(-((L+B)TP0)⊗Im--(P0(L+B))⊗Im-((L+B)TP0)⊗Im)S+2ST((P0(L+B))⊗Im)e≤-λmin(Q)∥S∥2+2∥S∥σmax(P0)σmax(L+B)∥e∥;V˙1 is negative as long as
(47)∥S∥>2σmax(P0)σmax(L+B)λmin(Q)∥e∥.
In light of the fact that
(48)∥qi-q0∥≤∥S∥σmin(L+B),
we have
(49)∥qi-q0∥≤2σmax(P0)σmax(L+B)λmin(Q)σmin(L+B)2V(0)Mme-c0t+2d0c0Mm.
Moreover,
(50)∥q˙i-q˙0∥≤∥ei∥+∥si∥≤(1+2σmax(P0)σmax(L+B)λmin(Q))×2V(0)Mme-c0t+2d0c0Mm.
From (49) and (50), we can see that the position and velocity tracking errors are cooperatively UUB and the ultimate bounds can be made as small as possible by appropriately tuning the design parameters.
Remark 12.
The work in [11] requires that the communication topologies among the followers are undirected. We extend these results to the case that the communication graph among the followers can be a directed graph, which is a more general case. The works in [22–24] considered the centralized fuzzy logic systems or the centralized tracking problem. Different from these works [22–24], we considered the decentralized systems, where the controller design and the systems analysis are more involved.
3.3. Fault-Tolerant Controller Design under Actuator Faults
From the above analysis, the proposed control law in (28) can guarantee arbitrary small tracking errors for the closed-loop system with fault-free actuators. This controller is called the nominal controller and denoted by τnor(i). To perform tracking control with actuator faults as defined by (10), an auxiliary controller τaux(i) is developed in addition to the nominal controller τnor(i), to compensate for the actuator faults. Consequently, the control input τi to the system shown in (10) is designed as
(51)τi=τnor(i)+τaux(i),
where τnor(i) is given by (28) and the fault-tolerant controller τaux(i) is synthesized as(52)τaux(i)=-(ki1tanh(ki1ei1/ςi1)σ-i1⋮kimtanh(kimeim/ςim)σ-im)-(τnor(i1)tanh(τnor(i1)ei1/υi1)σ-i1⋮τnor(im)tanh(τnor(im)eim/υim)σ-im),
with the positive design parameters ςij and υij, j=1,2,…,m.
Theorem 13.
Consider the networked Lagrange systems with the bias actuator faults and the partial loss of actuator effectiveness as defined by (10). Under the control law provided in (51) and the adaptive law given in (29), the cooperative tracking errors can be made as small as possible by suitably tuning the controller parameters.
Proof.
Substituting the fault-tolerant controller (51) into (25) gives
(53)Mie˙i+Ciei=(σiτi+fi+δi)+Fi(xi)-Miz˙i=τnor(i)+σiτaux(i)-ρiτnor(i)+fi+δi+Fi(xi)-Miz˙i,
where ρi=I-σi. The inequality 0≤∥ρi∥∞<1 always holds.
The candidate Lyapunov function V defined in Theorem 11 is also used here. Substituting (53) into the time derivative of V yields
(54)V˙=∑i=1n(eiTMie˙i+eiTCiei)-∑i=1n1βitr{θ~iTθ^˙i}=∑i=1neiT(τnor(i)+σiτaux(i)-ρiτnor(i)+fi+δi+Fi(xi)-Miz˙iτnor(i)+σiτaux(i)-ρiτnor(i))-∑i=1n1βitr{θ~iTθ^˙i}.
Applying the same analysis as that of Theorem 11, one has
(55)V˙≤-c0V+d0+∑i=1neiT(σiτaux(i)-ρiτnor(i)+fi)≤-c0V+d0+∑i=1n∑j=1meij(σijτaux(ij)-ρijτnor(ij)+fij).
In light of |ρij|<1 and Assumption 2, we have
(56)V˙≤-c0V+d0+∑i=1n∑j=1meijσijτaux(ij)+∑i=1n∑j=1m|eij||τnor(ij)|+∑i=1n∑j=1m|eij||kij(t)|.
Furthermore,
(57)V˙≤-c0V+d0+∑i=1n∑j=1meijσijτaux(ij)+∑i=1n∑j=1meijτnor(ij)tanh(eijτnor(ij)υij)+∑i=1n∑j=1mυij+∑i=1n∑j=1meijkijtanh(eijkijςij)+∑i=1n∑j=1mςij.
In light of (52), we have
(58)V˙≤-c0V+d0+∑i=1n∑j=1meijσij×(-kijtanh(eijkij/ςij)σ-ij-τnor(ij)tanh(eijτnor(ij)/υij)σ-ij)+∑i=1n∑j=1meijτnor(ij)tanh(eijτnor(ij)υij)+∑i=1n∑j=1mυij+∑i=1n∑j=1meijkijtanh(eijkijςij)+∑i=1n∑j=1mςij.
In light of Assumption 2, we have
(59)V˙≤-c0V+d0+∑i=1n∑j=1mυij+∑i=1n∑j=1mςij≤-c0V+d1,
with d1=d0+∑i=1n∑j=1mςij+∑i=1n∑j=1mυij.
Following the same procedure as that of Theorem 11, we get that the local tracking errors are cooperatively UUB.
Remark 14.
The works in [1–12] do not consider the actuator faults. We consider the actuator faults and present a fault-tolerant control scheme. The proposed fault-tolerant controller consists of the nominal controller (28) and the auxiliary controller (52). Different from the works [16–19], the fault detection and diagnosis mechanism is not required here. Thus, the proposed algorithm reduces the computation burden and decreases the response time of the controller. Compared with the work in [20], the proposed control law does not use the sign function. It avoids the chattering phenomenon.
4. Simulation Results
In this section, we present the simulation results to examine the performance of the proposed control scheme. The networked systems considered here consist of four standard two-DOF manipulators interconnected on a directed graph containing a spanning tree (see Figure 1). The manipulator consists of two serially connected links. The dynamics are described by (9) with
(60)Mi(qi)=[p1+p2+2p3cos(qi2)p2+p3cos(qi2)p2+p3cos(qi2)p2],Ci(qi,q˙i)=[-p3q˙i2sin(qi2)-p3(q˙i1+q˙i2)sin(qi2)-p3q˙i1sin(qi2)0],Gi(qi)=[p4gcos(qi1)+p5gcos(qi1+qi2)p5gcos(qi1+qi2)],
where p=[p1,p2,p3,p4,p5]=[2.9,0.76,0.87,0.34,0.87].
The communication topology.
Assume that the dynamics of the manipulators are unknown. Fuzzy logic systems are used to approximate the unknown dynamics. Choose the fuzzy membership functions as
(61)μAi1(xi)=exp(-(xi+1)2),μAi2(xi)=exp(-xi2),μAi3(xi)=exp(-(xi-1)2),
where xi is a variable which denotes the position or velocity of the manipulator.
The design parameters for the controller and adaptive laws are chosen as Ki=[800040], βi=500, ηi=0.01, and μi=1. The initial conditions are set as qi(0)=[00], q˙i(0)=[00], and θ^i(0)=0.
Case 1: Healthy Actuators. This case simulates the ideal situation when no actuator faults occur. Figure 2 provides the simulation results for the nominal controller (NC) given by (28). Figures 2(c) and 2(d) show the finite-time convergence of the velocity observers. Figures 2(a) and 2(b) demonstrate that the high control precision and an acceptable system performance have been achieved.
Simulation results with healthy actuators.
The desired and actual position trajectories of the manipulators
The desired and actual velocity trajectories of the manipulators
The estimated values for q˙0 at joint 1
The estimated values for q˙0 at joint 2
The control inputs of each manipulator
Case 2: Faulty Actuators. This example represents the severe case in which both the bias faults and the partial loss of actuator effectiveness occur. Assume that the actuators of the networked systems undergo a partial loss of effectiveness and a bias fault at t=10s. Let σ-i=0.02, kij(t)=2, and υij=ςij=0.5(i=1,2,3,4,j=1,2). In the simulation, the partial loss coefficient and the bias faults are set as
(62)σim={1t<10s0.15+0.1sin(0.5t)t≥10s(m=1,2),fim={0t<10s1+0.5sin(0.2t)t≥10s(m=1,2),
respectively.
Figures 3 and 4 show the results using the two different control laws based on the same simulation conditions. From Figures 3(a) and 3(b), we see that the effects of the actuator faults propagate to the Lagrangian dynamics and cannot be compensated by the nominal controllers. Significant degradation of the control performance and system instability after the faults can be observed. In contrast, Figures 4(a) and 4(b) show that the FTC controller can achieve the objective of fault-tolerant control and does succeed in compensating for the fault simultaneously.
Simulation results with the actuator faults under the nominal controller.
The desired and actual position trajectories of the manipulators
The desired and actual velocity trajectories of the manipulators
The control inputs of each manipulator
Simulation results with the actuator faults under the FTC.
The desired and actual position trajectories of the manipulators
The desired and actual velocity trajectories of the manipulators
The control inputs of each manipulator
In summary, for the healthy case, the proposed controller achieves the desired control performance of the closed-loop Lagrange system. For the cases with actuator faults, the proposed method can compensate for the effects of the faults. As the faults become more severe, the proposed controller still guarantees system stability.
5. Conclusion
A distributed fault-tolerant tracking control scheme was presented for networked uncertain Euler-Lagrange systems which may experience the bias faults and the partial loss of actuator effectiveness. Using finite-time observer based method and Lyapunov analysis we have derived stable adaptive protocol which consists of the nominal control input and the auxiliary control input, where the former achieved the distributed tracking control of healthy systems while the latter compensated for the effect of actuator faults. The proposed scheme was suitable for the general directed communication topology. The simulation results presented in this paper showed that the control scheme can successfully handle the model uncertainties and the unknown actuators faults.
Conflict of Interests
The authors declare that there is no conflict of interests regarding the publication of this paper.
Acknowledgments
This work was supported by the National Natural Science Foundation of China (Grant no. 61273108), the Fundamental Research Funds for the Central Universities (106112013CDJZR175501), and the Scientific Research Foundation for the Returned Overseas Chinese Scholars, State Education Ministry.
VicsekT.CzirókA.Ben-JacobE.CohenI.ShochetO.Novel type of phase transition in a system of self-driven particles1995756, article 122610.1103/PhysRevLett.75.12262-s2.0-0001730998JadbabaieA.LinJ.MorseA. S.Coordination of groups of mobile autonomous agents using nearest neighbor rules2003486988100110.1109/TAC.2003.812781MR19862662-s2.0-0038548185ChenG.LewisF. L.XieL.Finite-time distributed consensus via binary control protocols20114791962196810.1016/j.automatica.2011.05.013MR28868082-s2.0-80052029663MoreauL.Stability of multiagent systems with time-dependent communication links200550216918210.1109/TAC.2004.841888MR21164232-s2.0-12944293796Olfati-SaberR.MurrayR. M.Consensus problems in networks of agents with switching topology and time-delays20044991520153310.1109/TAC.2004.834113MR20869162-s2.0-4644244041RenW.BeardR. W.Consensus seeking in multiagent systems under dynamically changing interaction topologies200550565566110.1109/TAC.2005.846556MR21415682-s2.0-20344399896ChenG.LewisF. L.Robust consensus of multiple inertial agents with coupling delays and variable topologies201121666668510.1002/rnc.1615MR2809865ZBL1213.930092-s2.0-79952222066Rodriguez-AngelesA.NijmeijerH.Mutual synchronization of robots via estimated state feedback: a cooperative approach200412454255410.1109/TCST.2004.8250652-s2.0-4344690825RenW.Distributed leaderless consensus algorithms for networked Euler-Lagrange systems200982112137214910.1080/00207170902948027MR2561983ZBL1175.930742-s2.0-70349639398HouZ. G.ChengL.TanM.Decentralized robust adaptive control for the multiagent system consensus problem using neural networks200939363664710.1109/TSMCB.2008.20078102-s2.0-65549146417ChenG.YueY.SongY.Finite-time cooperative-tracking control for networked Euler-Lagrange systems20137111487149710.1049/iet-cta.2013.0205MR31150142-s2.0-84883315438ChenG.LewisF. L.Distributed adaptive tracking control for synchronization of unknown networked lagrangian systems201141380581610.1109/TSMCB.2010.20954972-s2.0-79957515933ŠiljakD. D.Reliable control using multiple control systems198031230332910.1080/00207178008961043MR5659252-s2.0-0018985792VeilletteR. J.MedanicJ. V.PerkinsW. R.Design of reliable control systems199237329030410.1109/9.119629MR11487122-s2.0-0026835057ZhaoQ.JiangJ.Reliable state feedback control systems designs against actuator failures199834101267127210.1016/S0005-1098(98)00072-72-s2.0-0032182773DingS. X.2008New York, NY, USASpringerMaY. J.JiangB.TaoG.ChengY. H.A direct adaptive actuator failure compensation scheme for satellite attitude control systems20142284542556ChenJ.PattonR. J.1999Boston, Mass, USAKluwer Academic PublishersDucardG.2009New York, NY, USASpringerXiaoB.HuQ.FriswellM. I.Robust fault tolerant control for spacecraft attitude stabilization under actuator faults and bounded disturbance20111335805100610.1115/1.4004061JiangB.GaoZ.ShiP.XuY.Adaptive fault-tolerant tracking control of near-space vehicle using TakagiSugeno fuzzy models20101851000100710.1109/TFUZZ.2010.20588082-s2.0-77957823980WangL. X.Fuzzy systems are universal approximatorsProceedings of the IEEE International Conference on Fuzzy SystemsMarch 1992San Diego, CA, USA116311702-s2.0-0026994365WangL.MendelJ. M.Fuzzy basis functions, universal approximation, and orthogonal least-squares learning19923580781410.1109/72.1590702-s2.0-0026928374SuX.WuL.ShiP.SongY.H∞ model reduction of Takagi-Sugeno fuzzy stochastic systems20124261574158510.1109/TSMCB.2012.21957232-s2.0-84869491170DanskinJ. M.The theory of max-min, with applications19661464166410.1137/0114053MR0210456